i have a webapp running on jboss-4.0.5.GA/apache-tomcat-5.5.20 that reads in 
the http post request body and processes it.

i noticed that for request bodies that didn't contain line separators and that 
had sizes that were exact multiples of 
org.apache.catalina.connector.CoyoteReader.MAX_LINE_LENGTH (4096), i was 
receiving null when calling org.apache.catalina.connector.CoyoteReader.readLine
().

i believe that the problem is at line 155 in 
org.apache.catalina.connector.CoyoteReader, where on the last iteration through 
the loop, "pos" does equal zero and null is returned even though data has been 
aggregated.

here's a command to run in cygwin to easily reproduce the problem:
for requestSize in 4095 4096 4097 8191 8192 8193; do dd if=/dev/zero bs=1c 
count=$requestSize | tr '\000' 'A' | curl --data-binary @- 
http://localhost:8080/DebugJboss/DebugServlet > $requestSize.txt; done;

output from directory listing (size filename):
4095 4095.txt
   0 4096.txt
4097 4097.txt
8191 8191.txt
   0 8192.txt
8193 8193.txt

here's the bulk of the servlet code i used to reproduce the problem:
public class DebugServlet extends HttpServlet {
	protected void doPost(HttpServletRequest arg0, HttpServletResponse arg1)
			throws ServletException, IOException {
		BufferedReader br = arg0.getReader();
		Writer writer = arg1.getWriter();
		String line = null;
		while ((line = br.readLine()) != null) {
			writer.write(line);
		}
		writer.close();
		br.close();
	}
}

it appears that a workaround is to wrap the requests's input stream instead:
BufferedReader br = new BufferedReader(new InputStreamReader(arg0.getInputStream
()));